Due to the interference effect of the existence of the isolated points and prone points, SVM-based segmentation algorithm cannot obtain an ideal segmentation effect.An image segmentation method based on visual attention and fuzzy SVM (MFSVM) with improved membership degree function is proposed.In order to avoid the interference from non-vital training samples and noises, the segmentation result is coincided with the characteristics of human vision.The new membership degree function not only considers the distance of samples to center, but also divides the sample points into two types according to the distance of sample points to the center.It has enhanced the effect of support vectors and can effectively improve the segmentation accuracy.Multiple sets of color images are selected to verify the effectiveness of the proposed method.Result shows that, comparing with the standard SVM and FSVM methods, the proposed method shows an effective segmentation result as well as good noise immunity ability.